Your team is deploying a new API on Cloud Run that utilizes an OpenTelemetry agent to export distributed tracing data to Cloud Trace for monitoring request latency. However, trace collection is inconsistent. How would you address this issue?